基于Landsat 8 OLI遥感影像和森林资源二类调查数据,对有林地、灌木林地、未成林地和非林地等林地类型,分别采用最大似然、神经网络、支持向量机和决策树分类方法进行分类,验证分类精度,并对分类效果进行对比评价。结果表明:支持向量机分类方法表现最好,分类精度为78.7%,Kappa系数为0.76;其次为神经网络和决策树分类方法,分类精度分别为76.8%和72.5%,Kappa系数分别为0.72和0.68;最大似然法表现最差,分类精度为44.9%,Kappa系数为0.39。研究结果可为森林资源信息的快速提取提供理论依据。

Based on Landsat-8 image and forest resources management survey data,different forest land types were classified by maximum likelihood classification (ML),neural net classification (NN),support vector machine classification (SVM) and decision tree classification (DT) methods,and then the precisions (P) of classifications were verified,and the performances of classifications were evaluated correlatively.The results show that the best performance was SVM (P=78.7%,Kappa=0.76),and the followings were NN (P=76.8%,Kappa=0.72) and DT (P=72.5%,Kappa=0.68),and the worst was ML (P=44.9%,Kappa=0.39).These results provide a theory basis for the rapid extraction of forest resources information of forestry science data platform.